On 4/18/2021 9:16 PM, Lars Kellogg-Stedman wrote:
If I have two virtual machines on two different hosts attached to
the same OVN virtual switch (so, communicating over Geneve tunnels),
is the correct MTU configuration:
(a) Lower the MTU inside the virtual machines, or
(b) Raise the MTU outside the virtual machines (e.g. on the host
physical interfaces)
Either will permit a (max size encapsulated frame + tunneling protocol
header) to traverse the local network. Changing the infrastructure
side (option b) seems at first glance like the easier solution, since
presumably the infrastructure is relatively static compared to the
virtual clients attached to the network, which may be dynamically
created and destroyed over time.
The problem with (b) is that it will result in > 1500 byte frames
leaving the local network, which maybe would be problematic? I'm not
sure, which is why I'm asking.
Looking at e.g [1] it looks like OpenStack went with option a (lower
the mtu inside the virtual machines), but that document doesn't really
explain the rational for the decision.
I think the answer ends up being "it depends". There are an almost
endless number of network configurations and it seems it would be hard
to have a 'one size fits all' solution in this space. If OpenStack
uses option (a) then the designers felt like that was the best for
their purposes but I've seen other customer networks where option
(b) was used as well.
- Greg
[1]:
https://docs.openstack.org/ocata/networking-guide/config-mtu.html#instance-network-interfaces-vifs
_______________________________________________
discuss mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
